Research Projects and Experiences

[P1] Integrated Traceability Analysis for Supporting Complex Safety Assessment of CPS Applications (March 2020~March  2021)

Key Responsibilities:  I was responsible for developing a tool for integrated traceability analysis of CPSs. I successfully validated the tool with a case study on platooning, rescue robots, and cooperative driving systems. I was also given the task of integrating our system with other collaborative work.  Six universities and over eight laboratories were involved in this project.

[P2] Learning-based safety analysis that supports intelligent CPS real-time collaboration (March 2021~March 2023)

Key Responsibilities:  I was responsible for developing a tool and simulation platform for learning-enabled safety analysis of CPS. I also served as the team leader in this project. I have developed a tool called CPSTracer to support learning-based safety analysis of CPSs and validated the tool with a case study on autonomous driving systems and collaborative robotics. I worked on a simulation platform for learning-based analysis, such as autonomous driving systems. We Analyzed the impact of extreme weather conditions and other environmental impacts on the safety of vision-based autonomous driving systems.

[P3] Safety Improvement of Intelligent Autonomous Systems based on Adversarial Deep Reinforcement Learning and its Quality Evaluation (July 2023~Present)

Key Responsibilities:  Currently, I am working as team lead for this project. I am working on adversarial attacks and defense methods for deep learning models used in safety-critical applications like autonomous driving. I am also working on tool development that can be used to generate adversarial scenario generation to analyze and improve the safety of autonomous driving systems. My team and I worked on developing adversarial defense methods such as adversarial attack detection, input transformation-based adversarial defense, and efficient adversarial training methods.

Developed Tools:

[T1] CPSTracer, A tool for composite hazard analysis of collaborative cyber-physical systems. We integrated the different hazard analysis techniques in Fault Tree Analysis, Failure Mode and Effect Analysis, and Event Tree Analysis in a single tool.  

[T2] TARDeep, An adversarial robustness testing tool for ADS models. I have recently developed a tool to analyze adversarial attacks and defense methods on autonomous driving systems and improve adversarial robustness. 

Honors and Awards: